The number
of items a company sells at any given location usually varies in a semi-predictable
way, based on the day of the week, time of the year, demand for the product, weather conditions, production and supply capabilities,
and many other factors.
As a simple example, a movie theater may sell more tickets when it
rains and on
Friday and
Saturday nights.
It may be the case that
rainy Saturdays increase ticket sales whereas
rainy weekdays decrease sales. This
type of relationship is called an
Interaction. Example:
Daily wind gust speed |
Whether or not it was a weekend |
The interaction between wind gust speed and weekends |
The resulting interaction is obtained by multiplying the two original variables.
In this case, the interaction helps us isolate the effect of wind gusts on weekend sales, while ignoring the
effect of wind gusts on sales during the week. |
Ticket sales may
increase when the
weather is
very hot (as people take advantage of your air-conditioned theater),
decrease when the temperature is
pleasant, and
increase again when
the weather is
extremely cold. This is known as a
Non-linear relationship between two variables.
Exploratory analysis can reveal which variables may be associated with
higher or lower ticket sales.
Some well-known modeling techniques include nonlinear multiple regression, binary
/ multinomial logistic regression, structural equation modeling, and canonical analysis. The technique used depends
on the type of data analyzed and the nature of the question posed.
Predictive models must be developed on subsets of data and then cross-validated
to ensure the model can estimate future sales.